Skip to content

Dataset · TIGER-Lab

Math instructions with CoT and PoT

TIGER-Lab/MathInstruct

Math problems with reasoning chains and programs for training instruction models.

What's inside

MathInstruct is an instruction dataset for training math models, collected from 13 datasets with solution explanations; six of them were prepared by the authors. It combines chain-of-thought and program-of-thought reasoning and covers various sections of mathematics. Data is text-based, in English, volume ranges from 100k to 1M examples.

What you can do with it

  • Training a math LLM on SFT instructions
  • Fine-tuning LoRA for CoT and PoT solutions
  • Testing model on reasoning tasks
Tasks
Text generation
Languages
English
Size
100K–1M rows
Downloads this month
13.4K
Likes
321

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("TIGER-Lab/MathInstruct")